Embodied Intelligence Sees Paradigm Shift! EinsiaAI Releases EmbodiedEvo, Self-Evolving Coding Agent Climbs to the Summit of Embodied Robodojo
具身智能迎来范式跃迁！EinsiaAI发布EmbodiedEvo，自进化Coding Agent登顶具身珠峰Robodojo
The Navers Lab team at EinsiaAI has introduced EmbodiedEvo, a self-evolving Coding Agent that accumulates coding skills and textual experience through interaction, enabling the Agent to complete tasks in unseen scenarios. In the RoboDojo benchmark, EmbodiedEvo achieved an average score of 45.08 and a success rate of 38.95%, surpassing all VLA/WAM, Agent, and Agent+VLA methods, taking the top position in the Agent track.
EmbodiedEvo represents a novel approach to self-evolving coding agents for embodied intelligence, enabling agents to accumulate and reuse experience across tasks and scenarios. It improves task success rates through iterative exploration and experience transfer, offering a new paradigm for training agents in physical environments.
Source: Machine Heart — Robotics on WeChat · Read original article ↗
English translation unavailable. Choose the original article to read it.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
37.82%
Reported success rate
69.67%
Reported success rate
53%
Reported success rate
100%
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
在被誉为 “具身智能界的珠穆朗玛峰” 的 RoboDojo 评测中，EmbodiedEvo 取得 45.08 的平均分、38.95% 的成功率，超过所有 VLA/WAM、Agent、Agent+VLA 的方法，登顶 Agent 赛道榜首。
Open source S9
在论文展示的本地评测中，EmbodiedEvo 取得 44.48 分、37.82% 成功率，超过表中全部对照方法。
Open source S20
其中 Memory 与 Open 成功率分别达到 69.67% 和 53.00%。
Open source S21
对齐积木和传送带匹配原本都有 60% 成功率，进化后分别达到 80%、100%。
Open source S23
堆叠与 Tic-tac-toe 的成功率从 0 提高至 100%，Imitation sorting 从 0 提高至 80%。
Open source S24
Source：Machine Heart — Robotics on WeChat · mp.weixin.qq.com